Unsupervised Investments (II): A Guide to AI Accelerators and Incubators
A list of 29 accelerators and incubators for AI startups
I. Rationale for the post
Well, let’s be completely honest: the current startups landscape is incredibly messy. Venture capitalist, angels, incubators, accelerators, private equity funds, corporate venture capital, private companies, research grants. There are plenty of ways to get funded to start your own company — but how many of them are not simply ‘dumb money’? How many of them give you some additional value and really help you scale your business?
This problem is particularly relevant for emerging exponential technologies such as artificial intelligence, machine learning and robotics. For those specific fields, highly specialized investors/advisors are essential for the success of the venture.
This is the reason why I wrote a long post on AI investors some time ago and why I am following up now with accelerators, which can be a valid investment alternative and business opportunity but that are commonly not fully understood.
But first, some fundamentals…
II. Who’s who in the funding game
Since the edges are blurring, it is hard to find a commonly shared definition for accelerators and incubators. Hence, I will provide two different definitions, one a bit more from a practitioner’s point of view, the other slightly more academic.
In the industry, the distinction between an accelerator and an incubator is simply related to the rationale for a company to join such a program. In other words, an incubator helps the entrepreneur in the development of her idea, while the accelerator focuses more on growing the business. The two programs have therefore two different goals and should be joined at a different stage of the startup lifecycle (Isabelle, 2013).
“a fixed-term, cohort-based program, including mentorship and educational components, that culminates in a public pitch event or demo day.”
From this definition is clear that the authors looked at different traits to characterize and distinguish different programs from each other. The key features can actually be summarized as follows.
Even though this academic definition clearly indicates thresholds and binary variables to identify different programs, it looks to me that — at least in the AI space — things are more complicated and actually it is really hard to define who is who (for help, check the brilliant review by Hausberg and Korreck, 2017). Furthermore, the important question we should ask is not whether to call a program accelerator or incubator, but rather what is the real value brought to the entrepreneur.
III. Are they worth their value?
If you are an entrepreneur, having so many different choices might make you wonder whether it might make sense to join one of those programs or not. And if you are an investor, a company, or anyone else looking at the space, you might start wondering if those programs suffer from an adverse selection problem: good companies go ahead with their feet while ‘lemon’ companies that cannot get funded or get the ball rolling go into these programs.
Entrepreneur Perspective: to join or not to join
Unless you are already an experienced entrepreneur, the short answer is yes, accelerators and incubators are worthy (Hallen et al., 2016). Starting and running a company is something no university can teach you (no matter how many innovation workshops you take or entrepreneurial courses you attend) but it is grounded on real life experience. In this respect, accelerator programs are sort of full-time educational bootcamp in which you rapidly learn what you need to at least survive the first year. Whether then you are gonna make it or not depends on how you transform that knowledge into the right actions.
Joining an accelerator is actually as reading a summary instead of the full book to do an exam: in this case, the full book would take you years to be read, while the summary takes a few months and can help you passing the exam. However, final graduation is a completely different thing.
‘Accelerators = Business synopsis’
Academic research, even if not unanimously (check this beautiful work by Yu, 2016), seems to confirm with data the value of those programs (Hochberg, 2015). Studies prove that accelerated companies reach milestones faster (Hallen et al., 2014), have a higher probability to raise further funding with respect to angel-supported startups (Winston-Smith and Hannigan, 2015), and that have even spillover effects on the entire entrepreneurial ecosystem (Fehder and Hochberg, 2015).
A warning though: even if some of those findings are true from a statistical point of view, there is a huge difference between different accelerators, and the quality of the program drastically impacts the positive effects for the startup.
Investor Perspective: should I stay or should I go
A good investor is basically the one who is able to:
i) pick straight the winner and helping him become bigger and stronger;
ii) pick a potential winner with the right things in place and helping him become successful.
The first case requires a lot of ex-ante work (due diligence) but not much after you invest. You simply seat down, relax and wait (it is not that simple actually, but let me go with this narrative for a second). The problem here is that there are few companies with these traits and everyone wants to invest in them, which considerably reduces the risk-return tradeoff.
The second case is instead more interesting and shows the real skills and contribution of the investor. It is also what it happens, most of the time and with exceptions, with companies coming out from accelerators and incubators program. These are companies that, for whatever reasons (lack of previous experience, no access to funding, etc.), might not have made it by themselves but are now in the game. Think of big success stories as Dropbox, for example.
So the question is: as an investor, should I invest in companies coming out from accelerator programs? Or am I buying a lemon?
The answer is ‘simple’, once again: yes, but mainly in those ones coming from excellent successful programs.
The proliferation of accelerators and incubators program made really difficult for investors to find real value in accelerated companies, especially for AI-related technologies and businesses. Good companies join accelerators for learning, mentoring and to get more exposure, all things as an entrepreneur you want to get from the best ones out there. And if good companies join an accelerator, the accelerator becomes more successful and attract better and better companies and founders on the next batches. It is a virtuous circle, which is creating a clear polarization in the industry, a positive skew distribution where very few programs deliver excellent results while the majority of them do not add any value (and in some cases are even detrimental) to the participants. In other words, I think there is a strong adverse selection problem in the accelerators/incubators space.
Of course, this is not a law of nature and does not imply that every company coming out from Techstars is going to become a unicorn (or the other way round). It is simply a rule of thumb to allocate a bit more efficiently your capital. If you are then able to spot out a potential winner in a low-level accelerator, chapeau, give yourself a pat on the shoulder because you did a very nice job.
Accelerators Assessment Metrics: is the program any good?
The common denominator of the two perspectives is that everything comes back to how good an acceleration program is. I have no particular experience in setting up or participating in an accelerator, so I do not know for sure the problems or the metrics on how to assess it. This is my interpretation (quite general with some sprinkle of AI somewhere), but feel free to comment below and tell me more about different metrics and aspects I should also consider:
i) Alumni network: who are the alumni of the program? This base represents the ‘customer base’ of the accelerator, so check it out if includes big names. Do not be trapped by average valuations of the portfolio of the program: having one Dropbox and dozen of ‘John Doe startups’ does not make it a good accelerator, it simply makes it a lucky one (look at different stats, if you want to, e.g., median, variance, etc.);
ii) Raising the next round: even though raising funds is not always a proof of business success, it is very often a good proxy for it. The more companies raise a further fund after the program, the better the program is;
iii) Raising a good next round: same considerations as above, with the additional aspect that companies need to raise a specific amount of money. The more companies can reach their funding goal, the better the program is.
Be careful: evaluating an accelerator on the basis of the average amount of dollars raised is a huge mistake and only increments the already existing hype on AI;
iv) Survival rate: the accelerators are set to provide entrepreneurs with tools and network to survive for at least 12 months (this is my view). The higher number of companies are still operating after one year, the better the accelerator was;
v) Exit: ceteris paribus, if companies coming out from programs are obtaining higher valuation than their competitors, shortening the time-to-exit, or simply increasing the probability of an exit, it means that the accelerator did the job it was supposed to.
However, this point is controversial for at least two reasons: first, it is statistically hard to understand how an accelerator affects a final exit. Life is much more complicated than linking straight accelerator → higher exit, but if all the companies coming out from a specific program obtain higher valuations with respect to their peers, we know for sure that there is some endogeneity there, even if we might not be able to identify the specific factors that make a business more successful.
Second, it depends on your view about business and what it means starting a company. Real visionary entrepreneurs do not start a company to sell it — they start something as it should run forever. An exit is somehow a defeat for some of them (there are exceptions, e.g., DeepMind), but the reality is that this class of entrepreneurs is disappearing. People start business nowadays with the idea in mind to sell out in 5 years to a specific buyer, or to use the technology developed to increase the salary base from $150k (a normal salary in big tech companies in the US for an AI researcher) to $7M (average amount got from acqui-hire in AI and machine learning sector).
I am not saying this is wrong and this is certainly what an investor wants, but it can invalidate the ‘Exit’ metric as one variable to track for accelerators’ performance;
vi) Wider network: a good accelerator has top-level mentors and knows how to engage them to be effective. It also has people behind who can really understand AI technologies and can help entrepreneurs with latest developments in research, or partners that can provide datasets for feeding neural nets.
IV. List of AI Accelerators and Incubators
I then compiled a list as extensive as possible of every accelerator, incubator or program I read or bumped into over the past months. It looks like there are at least 29 of them:
- AI Nexus Lab (NY): an intensive program run by Future Labs (NYU) and ff Venture Capital. During the program, the startups can get access even to NYU AI faculty, which means for some lucky entrepreneurs to potentially have the chance to work along side with Yann LeCun. They have just announced their first cohort: Alpha Vertex, Behold.ai, Cambrian Intelligence, HelloVera, Klustera;
- Alexa Accelerator (Seattle): powered by Alexa Fund in collaboration with Techstars, this accelerator has the goal of advancing voice-powered technologies. As one of the Techstars programs, startups receive $100k of funding upon acceptance in convertible notes, as well as $20k in exchange for 6% of equity (with a ‘Equity Back Guarantee’ clause, which basically gives the founders that chance to lower up to zero Techstars’ equity position within three days from the end of the program). Historically, it seems that Techstars companies go on to average more than $2M raised after the program;
- Bosch DNA (Berlin): the Indo-German accelerator targets startups in different areas which uses enabling technologies such as deep learning, analytics, AI and machine learning to go from “Lab to Market”. The Nurture program lasts for 18 weeks: the first 3 weeks are dedicated to idea validation, a short 10-days bootcamp, and mentors meeting. Phase II is about 10 weeks mainly running through customer validation, while finally phase III concerns pitching preparation for final demo days. Usually 5 Indian and 5 German startups are selected;
- Botcamp (NY): run by Betaworks’ team (very good media investors), it is a program specifically designed for conversational interfaces. A $200,000 uncapped, safe note with a 25% discount is offered to companies;
- Comet Labs (Bay area): I have already mentioned Comet Labs Research Team in a previous article on AI investors, but they are also product builders. They will run different ‘labs’ starting from this April. The first one just announced is the Transportation Lab, with two more to follow. The first cohort includes 7 (impressive to me) companies: Nomoko; AutoX; Oculii; Deep Vision; Minds.ai; Point One; Syntouch. They do not provide an investment by default but rather on a case by case basis (in the form of a warrant, a convertible note, or a discounted equity investment);
- Creative Destruction Lab (Toronto): this is a program longer than usual, but aimed to support entrepreneurs with an MVP with mentorship on how to raise a round, develop the go-to-market strategy and deal with legal, accounting, and other business processes;
- CyberLaunch (Atlanta): accelerator coming out from Georgia tech scene and with a focus on machine learning and information security. It is Chris Klaus’ second accelerator after Neurolaunch (focusing on neuroscience startups). They have incubated companies like C3Security, Chincapi, Cyberdot, Diascan, iTreatMD, Realfactor.io, Securolytics, Vyrill and Yaxa;
- Data Elite Ventures (Bay area): Tasso Argyros and Stamos Venios founded DEV in 2013 with the idea of accelerating and investing in big data companies. They look to be inactive for a while (or at least off the radar), despite having supported good companies (Unravel, Weft, 451 Degrees) and an exit done (Weft has been acquired by Genscape last year);
- Deep Science Ventures (London): DeepScienceVentures is not a proper AI oriented accelerator, but rather a deep tech lab where to incubate ideas. It targets people rather than companies, as you can notice from their cohort (very similar to what EF is doing). As a scientist, you join the DSV team for a 3-months internship and if you find the right idea and co-founders, you get access to the following 3-months of MVP prototyping;
- Element AI (Montreal): created by famous AI scientist Yoshua Bengio, JS Cournoyer, Jean-Francois Gagné, Nicolas Chapados this lab lies on the idea the Canadian AI ecosystem is still one of the strongest worldwide — and this is very true about talents as well as funding raised. It is a mixed between a pure research lab and an incubator, and it has been backed up by Real Ventures. It has been announced not more than a few months ago (although they got already funded by Microsoft Ventures), so there are no more precise information about how it will work in practice (except that they are already working on 10 different projects). Very recent news: they acquired the entire team at MLDB.ai, an open source machine learning database;
- Eonify (Los Angeles): they focus on healthcare vertical, so they offer perks such as help for Protocol development, regulatory applications, clinical trial design, or grant writing. There is not much more info out there about their accelerator program unfortunately;
- Founders Factory (London): the Factory is a much wider accelerator who happens to have though a specific track for AI companies. The idea seems to be co-creation/development of two-three AI businesses within the acceleration program every year, for five years. The first two companies, recently announced, are Iris.ai (science research assistant) and Illumr (organizational pattern detection);
- H2 Ventures (Sydney): H2 Ventures is an Australian venture capital specialized in fintech which will be running a first accelerator program for AI and data analytics companies starting next August. They have a few requirements (e.g., founding team no larger than 4 people) and they are likely the only Australian accelerator for AI startup. Applicants will need to demonstrate their ability to deliver an MVP within 6 months and the intention of raising a Series A round of capital within 6–12 months;
- IBM Alphazone (Israel): IBM created this accelerator with the goal in mind of fostering long-term technology and business partnerships with smaller companies in the Cloud, Big Data & Analytics and IoT space. They have another partnership in place with Becton, Dickinson and Company to jointly select up to 3 startups in healthcare delivery and decision making. For those startups they offer extra professional mentorship and matter experts, as well as a grant of up to $25,000. They supported NeuroApplied, Magentiq Eye and Articoolo;
- Innovat8 Connect (Singapore): a program that brings startups to work along side with Singtel group to develop new solutions useful to the group itself. Singtel Innov8, the VC arm of group (fund size of $250M) follows up with investments where and if needed. A good example of the program output is Xjera;
- Kapsch Factory1 (Vienna): The Factory1 Kapsch TrafficCom Accelerator 2017 is an acceleration program with a focus on future intelligent mobility solutions (Connected & Autonomous Driving, Big Data Analytics & Deep Learning, Smart Mobility). The CEO and a second team member (preferably the CTO) will have to be present in Vienna for the Kick-Off Bootcamp, the three Acceleration Weeks in Vienna and Berlin and the Demo Day in Montréal (Canada). All travel and accommodation costs are covered;
- Merantix (Berlin): run by Rasmus Rother (co-founder with Adrian Locher), Merantix is a venture builder specialized in AI and with a stronger focus on four specific verticals: Finance, Healthcare, Advertising and Automotive. Active since one year, they contributed to build companies like Blinq;
- Microsoft Accelerator (Bangalore): this accelerator program is within for a different reason. It has not been set up, to my knowledge, as an AI-accelerator, but though in the last cohort all the 14 companies accepted were doing some sort of AI/machine learning. In other words, this is the first ‘ex-post AI’ accelerator, because it has been changing its own nature by the companies it selected;
- NextAI (Toronto): a Canadian accelerator for startups with no previous funding. You can apply either as individual as well as a team (but first always apply as individual). It provides startups with a capital of 50k CAD with can be increased by a 30k as well as other 150k throughout the program for top performing teams incorporating a venture ($50,000 for a SAFE with a $2mm CAP and up to an additional $150,000 no CAP, 20% discount to next round). They also provide structured business and technical curriculum taught by successful entrepreneurs and award winning faculty from Rotman (University of Toronto), Harvard, MIT, NYU, and others;
- Nvidia Inception (Virtual): this is a virtual accelerator program that helps startups during product development, prototyping, and deployment. They can apply for GPU hardware grants and the NVIDIA Deep Learning Institute (DLI) will show the latest techniques in designing, training, and deploy neural network-powered machine learning in different applications. With respect to others, it looks like a soft program, but it directly makes startups to be considered for the GPU Ventures Program ($500K — $5M, and help in sales & marketing, joint development, and product distribution). Apparently, the Inception program includes over 1,300 startups up to date. 14 of those companies have been recently asked to pitch in front of investors and 6 of them eventually got funded through the venture programs (Abeja; Datalogue; Optimus Ride; SoundHound; TempoQuest; Zebra Medical);
- Play Labs (Cambridge, MA): this is a brand new accelerator, apparently only for MIT students and alumni. They have a strong focus on gamification and ‘playful technologies’, and provide companies with $20k funding plus other $80k (typically in convertible notes) at the end if certain requirements are met;
- Rockstart AI Accelerator (Netherlands): usually these guys run 5–6 months accelerators in Netherlands. The new program in AI is starting accepting applications in May and it will cost 6% of equity to startups (but only after having raised a further round of funding);
- Startup Garage (Facebook) (Paris): another brand new accelerator sponsored by Facebook within the startup campus called Station F. Facebook will provide 80 desks and space for 10–15 data-driven startups fro 6 months at no cost (or obligations to use FB products), as well as operational mentoring (marketing, legal, etc.) and technical help (from FAIR — Facebook AI Research). This confirms Facebook’s strategy to have a stronger technical presence in Europe and the ability of France to potentially become one of the major AI hub worldwide. According to VentureBeat, they have already selected a few startups for the first incoming program (Chekk; Mapstr; The Fabulous; Onecub; Karos);
- TechCode Global AI+ (Bay area): TechCode is a global network of startup incubators and entrepreneur ecosystems which will especially help companies in approaching the Chinese and Asian markets. 10 startups out of the 50 they selected for the program will benefit from an initial investment of $50k. Originally, they would earn a ‘success fee and equity stake’ only if the startup raised funding within 12months from the end of the program. Not sure how this changed for the Global AI+ program;
- The Hive (Bay area): they define it as a ‘co-creation studio to build and launch startups’ in AI (subdivided in deep learning, blockchain, AR, ‘ambient intelligence’ and ‘context computing’). They built companies as Sensify, Snips and Skry with their $22M second fund. They also host a meetup called ‘The Hive Think Tank’. Their business model is a bit atypical but not completely new: simply speaking, they either incubate existing companies or they think the idea, create the MVP and recruit executives to run this new startup;
- Voicecamp (Betaworks) (NY): as Botcamp above, this is also run by Patrick Montague and the Betaworks’ team but focuses on early stage companies building voice-based products. $200k uncapped, SAFE note with a 25% discount is offered to all the startups accepted.
- Winton Labs (London): the famous hedge-fund is now presenting the second cohort of its data science accelerator. First of all, it is really interesting to me that an investment firm in London decides to start an accelerator program without asking for anything in return. But it is more interesting to see what areas they want startups to work on: machine intelligence, forecasting, innovative data, or wildcard (not clear projects). Startups also get direct exposure to Winton Ventures, of course;
- Y Combinator (Bay area): Y Combinator is known to be one of (if not the) best accelerators in the world. They didn’t have any specific focus on AI until now, but they just announced an experimental batch on artificial intelligence. They claim to be agnostic to the industry and would eventually like to fund an AI company in every vertical. A specific thing they are looking for though is Robot Factories, and teams that use deep (reinforcement) learning to help to fix it.
- Zeroth AI (Hong Kong): Zeroth.AI is run by tak_lo and his team in Hong Kong, and has a wide spectrum of AI advisors although its young age and 10 early stage AI startups in their first cohort (4 of which in the bots/assistant space). This is probably going to change, with up to 20 startups and optional $120k of funding. The relocation for the program is not mandatory for the entire time frame but highly recommended at the beginning and at the end of the program.
I think it would be worthy to mention two other accelerators that focus on hardware but that, although not AI-focused, for the current historical moment we live in are incredibly close to the AI development: Industrio (Italy), a pure hardware accelerator, and Buildit (Estonia), an ‘accelerator of Things’.
BONUS PARAGRAPH: 10 Main Research Institutes
This is not really related to AI accelerators but I think worth to mention it for people working in the space. Some of the following institutes gather the best minds working on AI problems, and it might be useful for research developments, talent pipeline, as well as potential partnerships to keep track of them. I will not include in the following list the pure academic research institutes (i.e., the ones strictly belonging to/located within universities) because the list would be too long otherwise, and I won’t consider big tech companies as DeepMind, Google (Google Brain), Facebook (FAIR), Baidu, IBM, Microsoft and Toyota (but for an interesting discussion on the topic check here), as well as private research companies (e.g., Numenta, GoodAI, Cogitai, etc.). In no particular order then:
- The Alan Turing Institute (London);
- The Allen Institute for AI (Seattle);
- AI Research Institute (Korea);
- Machine Intelligence Research Institute (Berkeley, CA);
- Dalle Molle Institute — Swiss AI Lab (Manno, Switzerland);
- Sino-Israeli Robotics Institute (Guangzhou, China);
- The Montreal Institute for Learning Algorithms (MILA) (Montreal);
- OpenAI (San Francisco);
- Vector Institute (Toronto);
- Fondazione Bruno Kessler (Trento, Italy);
V. Final Food for Thoughts
I tried to list all the accelerators I could find working specifically on AI, and I hope it will help someone out there. It looks clear to me now that
i) the on-going confusion between accelerators and incubators facilitated the creation of mixed structures which have characteristics of both the programs;
ii) quality matters (not all the accelerator are equals). You get different value from different ecosystems even if the offer is the same on paper. Joining an accelerator in this list is also not a guarantee of success, and of course, there are many other excellent programs worldwide that can maybe work much better than some of the ones I showed above.
The motif, though (and my personal believe at this stage of AI development), is that specialized investors and accelerators can do a much better job in understanding and helping companies leveraging these exponential technologies.
There is also something else emerging from the list: there are really few AI accelerators/incubators in Silicon Valley proportionally speaking, although the common expectation would be to find most of them in the American entrepreneurial district.
My guess is that, in reality, from a pure cost-benefit perspective, the Bay Area is not the best place to start a company.
It is the best place though to expose the startup to a larger market, investors and public acknowledgement.
This does not imply that being in Silicon Valley makes no sense, but rather the opposite. I actually see shaping an emerging pattern in Silicon Valley, the same one that characterized in the past 30 years the pharmaceutical and movie industries. The pharma industry, for example, moved from being a large industry where the same company did the research (expensive), developed the molecules (expensive) and eventually commercialized the final product (cheap and with good margins), into a two-ways sector where biotech companies took the higher risk of developing experimental molecules while big pharma corporations were in charge of FDA regulation approval and market launch.
Of course, it is a bit more complicated than that, but the main message is that the sector self-specialized and assigned to each class of players what they knew how to do more efficiently (research for biotech and commercialization for pharma companies).
In the same way, it will make sense probably to develop companies in other countries (where the real cost of starting up is much lower) to eventually land in California only once ready to either scale, raise larger rounds of financing or massively go to market.
A final interesting thing I noticed, which might be useful to some entrepreneurs: it is coming out the new concept of ‘specialized co-working space’, and we have something focusing on AI called RobotX Space in multiple cities (Silicon Valley and Asia). I have never been there (but hopefully I will in the future) but I think that it makes a lot of sense to create technology hubs like this one. This model might, in the future, even undermine the business models of accelerators and incubators.
As I always say, this type of list is the result of an intensive research work on publicly available data, but it can be still prone to errors or lacks. So, if I misled something or forgot someone, got in touch and let me know!
Cohen, S. (2013). “What Do Accelerators Do? Insights from Incubators and Angels”. Innovations, 8:3/4: 19–25.
Cohen, S., Hochberg, Y. V. (2014). “Accelerating Startups: The Seed Accelerator Phenomenon”. Working paper.
Fehder, D. C., Hochberg, Y. V. (2014). “Accelerators and the Regional Supply of Venture Capital Investment”. Working paper.
Hallen, B. L., Bingham, C., Cohen, S. (2014). “Do Accelerators Accelerate? A Study of Venture Accelerators as a Path to Success”. Academy of Management Annual Meeting Proceedings.
Hallen, B. L., Bingham, C., Cohen, S. (2016). “Do Accelerators Accelerate? The Role of Indirect Learning in New Venture Development”. Available at SSRN: https://ssrn.com/abstract=2719810
Hausberg, J. P., Korreck, S. (2017). “A Systematic Review and Research Agenda on Incubators and Accelerators”. Available at SSRN: https://ssrn.com/abstract=2919340.
Hochberg, Y. V. (2015), “Accelerating Entrepreneurs and Ecosystems: The Seed Accelerator Model,” in Innovation Policy and the Economy, Volume 16, Josh Lerner and Scott Stern editors, National Bureau of Economic Research.
Isabelle, D. A. (2013). “Key Factors Affecting a Technology Entrepreneur’s Choice of Incubator or Accelerator”. Technology Innovation Management Review: 16–22.
Kim, J. H., Wagman, L. (2014). “Portfolio size and information disclosure: An analysis of startup accelerators”. Journal of Corporate Finance, 29: 520–534.
Yu, S., (2016). “How Do Accelerators Impact the Performance of High-Technology Ventures?”. Available at SSRN: https://ssrn.com/abstract=2503510
Winston-Smith, S., Hannigan, T. J. (2015). “Swinging for the fences: How do top accelerators impact the trajectories of new ventures?”. Working paper.
Look at my other articles on AI and Machine Learning: